Math/programming question, but it's related to MARP! I promise!

greenspun.com : LUSENET : MAME Action Replay : One Thread

WARNINGS: BIIIIIIIIIIG POST! I'm sorry! I just had a lot to say!

OK. Today, I was trying to set a high score on "Eliminator" and I had an idea to write a program which would accept a chronological list of scores, and it would tell you a few things:

* How consistently you play (What percentage of your score is not random?) * How many more games you have expect to play to achieve a certain score (an upper and lower bound, based on your consistency and how much you're improving)

Anyways. I've got the program pretty much working. It even works for tracking multiple games in a single input file. It's neat. But I hit two small snags.

First of all, I'm using Linear Regression to predict your future scores. This is bad, because in general people's scores either increase in a quadratic fashion (Galaga or Mr. Do for example) or in a logarithmic fashion (Ms. Pac Man, for example). So if anyone has any ideas for how to implement Quadratic Regression let me know. Also, if anyone has any ideas for how to mathematically detect for logarithmic regression, please let me know also!

Okay! Second snag! Basically, I plugged my data for Eliminator into the program and... Well, it turns out I have this problem where if I play a game 20 times in one sitting, my last scores will be worse than my earlier scores. I think it's a combination of stress and boredom, and I've noticed it happen when playing other non-arcade games, too. But in any case, I was curious if anybody had an idea for how to combat this? My only thought was to maybe allow the user to separate groups of recordings with a row of asterisks, or perhaps a date. That way, maybe you can still predict, "You will reach your goal in (n) days at this rate..." Only problem is how to make this prediction mathematically, without throwing out most of the data! Any suggestions?

OK! Well like I said, I have the prototype program done, though there may still be bugs. If anybody wants the source let me know. Also, if you trust me, or your virus-checking software enough to let me send a compiled .EXE over, that's fine too. Don't expect too much as far as documentation or "zazz", because it's still in pre-beta stages.

- Aaron Pieper

-- Aaron Pieper (apieper@vt.edu), December 18, 2000

Answers

First snag can be solved by using and decent statistical analysis package (you don't have to limit yourself to quadratics, you can go cubes/logs/powers etc but it will take longer to figure the function for complex modeling.) don't bother reinventing the wheel for this stuff, there's a lot of free statistical stuff out there, i think you could even do it in Excell. there is even a site where you can perform quadratic regression on line: http://www.shodor.org/succeed/compchem/tools/quadfit.html

second snag is impossible, since scores don't increase continually by times played unless it's Hisa but even he has limits. the only reason galaga scores appear to increase quadratically is beause eventually you can reach a point where the difficulty is the same and you can repeat what you've done in a past level and score infinite amounts dependent on boredom/stamina or sore apendages.

-- Chad (churritz@cts.com), December 18, 2000.


Yes, your idea is the tops! I will try doing this in Excel. Although reinventing the wheel sounds like fun, too.

-- Aaron Pieper (apieper@vt.edu), December 18, 2000.

I don't see second snag as a snag at all. That's the way with statistical data. Even if your scores drops at the end of a session, it will (hopefully) increase over time. After a few sessions the regression will find the correct trends and you can get a descent approximation on how many days you have to play before reaching your goal.

-- Abbe (Albin@518.nu), December 19, 2000.

OK. very good calculate man.
Simply I want to know the date of "ANY MARPER"over these "possible"scores.
Please predict them. SDI--------------79M+
ROMPERS---------11M+
Rod-Land(Seacret)--79M+
Do-donpachi------719M+
V-five------------36M+
ShockTroopers-----31M+
Quarth------------5M+
AfterburnerII-------324M+
TruxtonII----------10M+
Danganfeveron-----14M+
donpachi----------59M+
Esprade-----------40M+
Pacmania----------2800K+
galaga'88----------2170K+
street fightersII-----2130K+
exchanger(vs)------9,999,999
motos---2,651,300(perfect)


-- Rank-52 (weavuspert@yahoo.co.jp), December 19, 2000.

ok hisa, when do we get to see your Truxton 2 recording?

-- Richy Cream (urgod@eudoramail.com), December 19, 2000.


Hisa: I'm working on the Pacmania one, but you have to have it set up for an endless game then. I think 5M is doable on Pacmania - endless game, rank D. Should have it up by the end of 2002.

-- Matt Denham (peng@demongate.ath.cx), December 20, 2000.

Hey, yesterday I scored at galaga88 1000 points, today 2000, tomorrow gonna do 4000. How long will it take for me to score over 3 000000?

Seriously, which function do you use to forecast at Excel, sounds interesting...what if you play several games in a session, do you count only the best one or average,or?

Sounds cool way to see when there's no point to try improve a score, if you count that it takes too much time/effort for a minor improvement(So far I haven't seen a game where my score improvement would have totally stopped though).

If you try to maximize a score, it WOULD be wise to play the game every day to keep the touch. But sometimes a few days break does only good...

just my thoughts, Tommi

-- Tommi (tiihoto@hotmail.com), December 21, 2000.


Moderation questions? read the FAQ